Under Water Image Super-Resolution Using Generative Adversarial Networks With Spatial Attention Mechanism

Underwater imaging often encounters challenges such as low resolution and diminished clarity due to the effects of light absorption and scattering in aquatic environments. To address these issues, this study presents an enhanced image super-resolution method that integrates a Spatial Attention Module (SAM) within the ESRGAN generator architecture. The proposed model enables focused reconstruction of critical spatial features, such as edges and textures, which are commonly lost in traditional interpolation methods. Comparative evaluations against conventional upscaling techniques— namely nearest neighbor, bilinear, and bicubic interpolation— highlight the effectiveness of the approach. Experimental results demonstrate that the SAM-enhanced ESRGAN achieves a Peak Signal to-Noise Ratio (PSNR) of 28.53 dB and a Structural Similarity Index Measure (SSIM) of 0.821, marking a substantial improvement in both visual fidelity and quantitative accuracy over baseline methods.